This paper describes an experiment to obtain an upper bound on the accuracy that can be expected when attempting to synchronise the system clock of one computer with the system clock of a physically remote computer. We used one computer in the U.K. and a number of computers in the U.S., linked by a complex datagram network, to provide a worst-case environment. This paper gives some reasons for wanting to synchronise clocks and the relationship to previous work. The experimental problems are discussed and the results analysed.